Algorithms used for “time series trend”
1 min readJun 30, 2020
- Linear Regression:
pros:- Ability to handle different time series components and features.
cons:- Sensitive to outliers and strong assumptions. - Exponential Smoothing:
pros:- Ability to handle variable level,trend and seasonality components.Automated optimization.
cons:- Narrow confidence intervals and sensitive to outliers. - ARIMA(Auto-regressive integrated moving average):
pros:- High interpretability,realistic confidence intervals and unbaised forecasts.
cons:- requires more data,strong restrictions and assumptions and hard to automate. - Dynamic Linear Model:
pros:- More transparent than other models,deals well with uncertainty, control the variance of the components and high interpretability
cons:- Higher holdout error,higher training and evaluation time. - Neural Network Model:
pros:- Less restrictions and assumptions. Ability to handle complex non-Linear patterns,high predictive power and can be easily automated.
cons:- low interpretability,difficult to derive confidence intervals for the forecasts and requires more data.